From this paper, I see that the fidelity of single qubit gate after using dynamical decoupling only reach around 0.85 while I normally saw experiment papers state their fidelity can reach around 0.99 to be a good result. So my question is, does 0.85 a good fidelity with the method of dynamical decoupling, if so, I think this fidelity is a little bit low?
I think you're confusing this with reported numbers for average gate fidelities.
The authors are estimating the fidelity of the time-evolved state with the initial state $|\psi\rangle$. If you prepare a state $|\psi\rangle$, and not do anything, it will (among other things) decohere such that the fidelity of the time-evolved state with $|\psi\rangle$ decreases with time. In particular, it is not surprising that you don't find great fidelities if you wait long enough. The purpose of dynamical decoupling is to counter-act this, at least to some extent, and that's what's presented in this plot.
The thing is that DD is an error mitigation technique, so it does not need fully-fledged error correction to work. For certain NISQ applications, this might be enough. Besides, this can be combined with QEC, and less noise usually means less overhead by QEC!